A method for intelligent support to medical diagnosis in emergency cardiac care

Usually, for a patient to receive a cardiac examination, they need to move from their location to a nearby hospital, or to get an equipped ambulance to where they are situated. In emergency cases, specialists also have to be available for diagnosing and deciding on how to proceed with that patient,...

Full description

Saved in:
Bibliographic Details
Published in2017 International Joint Conference on Neural Networks (IJCNN) pp. 4587 - 4593
Main Authors Souto Maior Neto, Luis A., Pequeno, Robson, Almeida, Carlos, Galdino, Katia, Martins, Fabricia, de Moura, Antonio V.
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.05.2017
Subjects
cardiac arrhythmia
Discrete wavelet transforms
Electrocardiography
Electrodes
Feature extraction
Heart beat
pattern recognition
portable medical care
Online AccessGet full text

Cover

More Information
Summary:Usually, for a patient to receive a cardiac examination, they need to move from their location to a nearby hospital, or to get an equipped ambulance to where they are situated. In emergency cases, specialists also have to be available for diagnosing and deciding on how to proceed with that patient, given their condition. In many Brazilian inland cities, however, there is a lack of equipment or personnel able to perform such tasks. In this paper, we propose a system for assisting remote electrocardiographic (ECG) diagnosis. This system consists of performing exams using a portable ECG device, sending it to a cloud service accessible remotely through devices such as mobile phones, tablets, laptops or computers, where ECG data are processed by an intelligent arrhythmia detector (IDAH-ECG), which, upon detecting abnormal heartbeats, informs physicians in the area via Short Message Service (SMS). IDAH-ECG makes use of signal processing tools such as Finite Impulse Response filter, Stationary Wavelet Transform (SWT) filter, and Discrete Wavelet Transform (DWT) filter, along with DWT feature extraction and Principal Component Analysis dimensionality reduction for preprocessing. A Multilayer Perceptron Neural Network performs classification between normal and abnormal beats. The classifier has yielded an average accuracy of 96.48%, sensitivity of 98.70%, specificity of 94.45%, performed by both Monte Carlo and ten-fold cross validation techniques.
ISSN:2161-4407
DOI:10.1109/IJCNN.2017.7966438